TY - JOUR

T1 - Throughput Time Predictions Along the Order Fulfilment Process

AU - Hiller, Tobias

AU - Demke, Tabea Marie

AU - Nyhuis, Peter

PY - 2024/1/22

Y1 - 2024/1/22

N2 - Planned times for the throughput of production are key components to production planners for determining delivery dates with customers, capacity planning, scheduling and order coordination. While traditional estimation methods often rely on basic statistics and expert knowledge, data mining respectively machine learning offers the potential to compute more precise predictions for order-specific planned throughput times. Factors that lead to deviations from the plan are diverse and thus challenging to consider in the various production planning tasks along the order fulfilment process. Intelligent throughput time predictions promise a remedy. Yet, predictive models are often not designed to be practically applicable due to a lack of consideration of the various characteristics of each stage of the order fulfilment process. To address this gap, this paper takes a closer look at the prediction of throughput times for the various stages of order fulfilment. Based on the Cross Industry Standard Process for Data Mining, the characteristics of the individual steps to build a prediction model are elaborated with a focus and business and data understanding and then examined in a case study. From that, practical implications are derived and guidance for practitioners is given. A key finding is that predictions are less accurate in the early stages of order fulfillment. Prediction quality naturally enhances over time, since more and more order details are known. In conclusion, an iterative prediction process with an evolving database ensures good prediction quality, especially in the late stages of order fulfillment.

AB - Planned times for the throughput of production are key components to production planners for determining delivery dates with customers, capacity planning, scheduling and order coordination. While traditional estimation methods often rely on basic statistics and expert knowledge, data mining respectively machine learning offers the potential to compute more precise predictions for order-specific planned throughput times. Factors that lead to deviations from the plan are diverse and thus challenging to consider in the various production planning tasks along the order fulfilment process. Intelligent throughput time predictions promise a remedy. Yet, predictive models are often not designed to be practically applicable due to a lack of consideration of the various characteristics of each stage of the order fulfilment process. To address this gap, this paper takes a closer look at the prediction of throughput times for the various stages of order fulfilment. Based on the Cross Industry Standard Process for Data Mining, the characteristics of the individual steps to build a prediction model are elaborated with a focus and business and data understanding and then examined in a case study. From that, practical implications are derived and guidance for practitioners is given. A key finding is that predictions are less accurate in the early stages of order fulfillment. Prediction quality naturally enhances over time, since more and more order details are known. In conclusion, an iterative prediction process with an evolving database ensures good prediction quality, especially in the late stages of order fulfillment.

KW - Capacity planning

KW - data analysis

KW - Job shop scheduling

KW - machine learning

KW - Predictive models

KW - Procurement

KW - Production

KW - Production planning

KW - Task analysis

KW - Throughput

KW - throughput time prediction

UR - http://www.scopus.com/inward/record.url?scp=85182936125&partnerID=8YFLogxK

U2 - 10.1109/ACCESS.2024.3353029

DO - 10.1109/ACCESS.2024.3353029

M3 - Article

AN - SCOPUS:85182936125

VL - 12

SP - 9705

EP - 9718

JO - IEEE ACCESS

JF - IEEE ACCESS

SN - 2169-3536

ER -